Arterial blood gases, CBCs, and BMPs from a major hospital center
Description
Background: Spurious hypoxemia—artefactual reduction in measured arterial oxygen pressure (paO₂) from cellular oxygen consumption in blood specimens—has been attributed to leukocytosis (“leukocyte larceny”) or thrombocytosis (“platelet larceny”). However, the relative contribution of elevated cell counts and temperature-dependent metabolism remains unclear. Methods: We retrospectively analyzed 7,431 arterial blood gas (ABG) specimens from 2,643 patients at UC Davis Medical Centre (collection-to-analysis time ≤80 minutes). Spurious hypoxemia was defined paO₂ <60 mmHg (reported at the patient's body core temperature) with oxygen saturation ≥90% by co-oximetry. Kaplan–Meier analysis, Cox proportional hazards regression, and mixed-effects logistic regression assessed predictors. The institutional review board approved the study with waiver of consent (IRB ID:2065699-1). Results: Extreme leukocytosis (WBC ≥50×10⁹/L) occurred in 31 specimens (0.4%; median 65.4×10⁹/L) and thrombocytosis (platelets ≥500×10⁹/L) in 126 specimens (1.7%; median 576×10⁹/L). Spurious hypoxemia occurred in 176 specimens (2.4%). Each 1°C decrease in body temperature increased odds of spurious hypoxemia by 85% (OR 1.85, 95% CI 1.28 to 2.70, p=0.001). Each 1 mg/dL increase in serum creatinine increased the odds of spurious hypoxemia by 31% (OR 1.31, 95%CI 1.02 to 1.68, p=0.037); and each 0.01 unit increase in pH increased the odds of spurious hypoxemia by 12% (OR 1.12, 95% CI 1.07 to 1.16, p < 0.0001. Neither leukocytosis nor thrombocytosis predicted spurious hypoxemia when added to temperature-based models. Adding cell counts did not improve model discrimination (ΔAUC ≤0.001). Conclusions: Patient temperature at specimen collection, not extreme leukocytosis or thrombocytosis, is the primary determinant of spurious hypoxemia. These findings challenge the traditional “leukocyte larceny” paradigm and have implications for blood gas specimen handling in hypothermic patients.
Files
Steps to reproduce
We conducted a retrospective cohort study at a major California medical center using electronic health record–derived laboratory data (September 1, 2020–January 15, 2026). Inclusion required an arterial blood gas (ABG) with a paired complete blood count and basic metabolic panel within 24 h, plus co-oximetry. Missing data were handled by multiple imputation. The full dataset included 7,473 specimens from 2,647 patients; restricting the ABG to collection-to-analysis time ≤ 80 min yielded a final analytic cohort of 7,431 specimens from 2,643 patients. Outcome definitions The primary outcome was spurious hypoxemia (PaO₂ <60 mmHg reported at the patient's body core temperature, with co-oximetry sO₂ ≥90%). The secondary outcome was sO₂ discrepancy (absolute difference ≥5 percentage points between measured and expected sO₂). Expected sO₂ was calculated from PaO₂ standardized to 37°C and pH 7.40. Extreme leukocytosis was WBC ≥50×10⁹/L and thrombocytosis was platelets ≥500×10⁹/L; alternative age-based z-score definitions were available but not used. This spurious hypoxemia definition flags physiologically implausible combinations of low PaO₂ with preserved measured saturation, often seen as discordance between PaO₂-based estimates and co-oximetry. Because PaO₂ is reported at the patient’s core temperature (temperature-corrected), we also analyzed sO₂ discrepancy (measured vs expected) as a broader measure of discordance. Temperature stratification Temperature at collection was classified using age-specific thresholds for neonates, children, and adults; borderline ranges were evaluated in sensitivity analyses. Expected oxygen saturation: Expected sO₂ was computed using Severinghaus for standardized PaO₂ <200 mmHg and Roughton–Severinghaus for PaO₂ ≥200 mmHg, after standardizing PaO₂ to 37°C and pH 7.40. Statistical analysis Associations between processing time and WBC/platelet counts were assessed with Pearson correlations and mixed-effects linear models with patient-level random intercepts. Collection-to-analysis time was summarized with Kaplan–Meier curves and modeled with Cox regression using robust standard errors clustered by patient; for sparse/zero-event strata we used Firth penalized Cox regression. ROC and multivariable modeling Prespecified logistic regression models were compared using BIC and AUC (DeLong 95% CIs) to test whether adding WBC/platelet terms improved prediction of spurious hypoxemia or sO₂ discrepancy. Optimal cut-points used Youden’s J with bootstrap CIs. Primary analyses used mixed-effects logistic regression with a patient random intercept. Predictors for spurious hypoxemia were temperature (per °C), collection-to-analysis time, neonatal status, serum creatinine, and pH; predictors for sO₂ discrepancy were analysis time, neonatal status, age (≥2 months), serum creatinine, and pH. WBC and platelets were added as thresholds and as continuous covariates. R was used to analyze the data.
Institutions
- University of California, DavisCalifornia, Davis